From 933508e2a2f7923cebc15d679b78933df8fb9ee5 Mon Sep 17 00:00:00 2001 From: MerryMage Date: Thu, 3 Aug 2017 12:22:51 +0100 Subject: interpolate: Interpolate on a frame-by-frame basis --- src/audio_core/hle/source.cpp | 49 ++++++++++-------------- src/audio_core/interpolate.cpp | 86 +++++++++++++++++++----------------------- src/audio_core/interpolate.h | 27 +++++++------ 3 files changed, 74 insertions(+), 88 deletions(-) (limited to 'src/audio_core') diff --git a/src/audio_core/hle/source.cpp b/src/audio_core/hle/source.cpp index 92484c526..de4e88cae 100644 --- a/src/audio_core/hle/source.cpp +++ b/src/audio_core/hle/source.cpp @@ -244,17 +244,27 @@ void Source::GenerateFrame() { break; } - const size_t size_to_copy = - std::min(state.current_buffer.size(), current_frame.size() - frame_position); - - std::copy(state.current_buffer.begin(), state.current_buffer.begin() + size_to_copy, - current_frame.begin() + frame_position); - state.current_buffer.erase(state.current_buffer.begin(), - state.current_buffer.begin() + size_to_copy); - - frame_position += size_to_copy; - state.next_sample_number += static_cast(size_to_copy); + switch (state.interpolation_mode) { + case InterpolationMode::None: + AudioInterp::None(state.interp_state, state.current_buffer, state.rate_multiplier, + current_frame, frame_position); + break; + case InterpolationMode::Linear: + AudioInterp::Linear(state.interp_state, state.current_buffer, state.rate_multiplier, + current_frame, frame_position); + break; + case InterpolationMode::Polyphase: + // TODO(merry): Implement polyphase interpolation + LOG_DEBUG(Audio_DSP, "Polyphase interpolation unimplemented; falling back to linear"); + AudioInterp::Linear(state.interp_state, state.current_buffer, state.rate_multiplier, + current_frame, frame_position); + break; + default: + UNIMPLEMENTED(); + break; + } } + state.next_sample_number += frame_position; state.filters.ProcessFrame(current_frame); } @@ -305,25 +315,6 @@ bool Source::DequeueBuffer() { return true; } - switch (state.interpolation_mode) { - case InterpolationMode::None: - state.current_buffer = - AudioInterp::None(state.interp_state, state.current_buffer, state.rate_multiplier); - break; - case InterpolationMode::Linear: - state.current_buffer = - AudioInterp::Linear(state.interp_state, state.current_buffer, state.rate_multiplier); - break; - case InterpolationMode::Polyphase: - // TODO(merry): Implement polyphase interpolation - state.current_buffer = - AudioInterp::Linear(state.interp_state, state.current_buffer, state.rate_multiplier); - break; - default: - UNIMPLEMENTED(); - break; - } - // the first playthrough starts at play_position, loops start at the beginning of the buffer state.current_sample_number = (!buf.has_played) ? buf.play_position : 0; state.next_sample_number = state.current_sample_number; diff --git a/src/audio_core/interpolate.cpp b/src/audio_core/interpolate.cpp index 8a5d4181a..16e68bc5c 100644 --- a/src/audio_core/interpolate.cpp +++ b/src/audio_core/interpolate.cpp @@ -13,74 +13,64 @@ namespace AudioInterp { constexpr u64 scale_factor = 1 << 24; constexpr u64 scale_mask = scale_factor - 1; -/// Here we step over the input in steps of rate_multiplier, until we consume all of the input. +/// Here we step over the input in steps of rate, until we consume all of the input. /// Three adjacent samples are passed to fn each step. template -static StereoBuffer16 StepOverSamples(State& state, const StereoBuffer16& input, - float rate_multiplier, Function fn) { - ASSERT(rate_multiplier > 0); +static void StepOverSamples(State& state, StereoBuffer16& input, float rate, + DSP::HLE::StereoFrame16& output, size_t& outputi, Function fn) { + ASSERT(rate > 0); - if (input.size() < 2) - return {}; + if (input.empty()) + return; - StereoBuffer16 output; - output.reserve(static_cast(input.size() / rate_multiplier)); + input.insert(input.begin(), {state.xn2, state.xn1}); - u64 step_size = static_cast(rate_multiplier * scale_factor); + const u64 step_size = static_cast(rate * scale_factor); + u64 fposition = state.fposition; + size_t inputi = 0; - u64 fposition = 0; - const u64 max_fposition = input.size() * scale_factor; + while (outputi < output.size()) { + inputi = static_cast(fposition / scale_factor); - while (fposition < 1 * scale_factor) { - u64 fraction = fposition & scale_mask; - - output.push_back(fn(fraction, state.xn2, state.xn1, input[0])); - - fposition += step_size; - } - - while (fposition < 2 * scale_factor) { - u64 fraction = fposition & scale_mask; - - output.push_back(fn(fraction, state.xn1, input[0], input[1])); - - fposition += step_size; - } + if (inputi + 2 >= input.size()) { + inputi = input.size() - 2; + break; + } - while (fposition < max_fposition) { u64 fraction = fposition & scale_mask; - - size_t index = static_cast(fposition / scale_factor); - output.push_back(fn(fraction, input[index - 2], input[index - 1], input[index])); + output[outputi++] = fn(fraction, input[inputi], input[inputi + 1], input[inputi + 2]); fposition += step_size; } - state.xn2 = input[input.size() - 2]; - state.xn1 = input[input.size() - 1]; + state.xn2 = input[inputi]; + state.xn1 = input[inputi + 1]; + state.fposition = fposition - inputi * scale_factor; - return output; + input.erase(input.begin(), input.begin() + inputi + 2); } -StereoBuffer16 None(State& state, const StereoBuffer16& input, float rate_multiplier) { - return StepOverSamples( - state, input, rate_multiplier, +void None(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output, + size_t& outputi) { + StepOverSamples( + state, input, rate, output, outputi, [](u64 fraction, const auto& x0, const auto& x1, const auto& x2) { return x0; }); } -StereoBuffer16 Linear(State& state, const StereoBuffer16& input, float rate_multiplier) { +void Linear(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output, + size_t& outputi) { // Note on accuracy: Some values that this produces are +/- 1 from the actual firmware. - return StepOverSamples(state, input, rate_multiplier, - [](u64 fraction, const auto& x0, const auto& x1, const auto& x2) { - // This is a saturated subtraction. (Verified by black-box fuzzing.) - s64 delta0 = MathUtil::Clamp(x1[0] - x0[0], -32768, 32767); - s64 delta1 = MathUtil::Clamp(x1[1] - x0[1], -32768, 32767); - - return std::array{ - static_cast(x0[0] + fraction * delta0 / scale_factor), - static_cast(x0[1] + fraction * delta1 / scale_factor), - }; - }); + StepOverSamples(state, input, rate, output, outputi, + [](u64 fraction, const auto& x0, const auto& x1, const auto& x2) { + // This is a saturated subtraction. (Verified by black-box fuzzing.) + s64 delta0 = MathUtil::Clamp(x1[0] - x0[0], -32768, 32767); + s64 delta1 = MathUtil::Clamp(x1[1] - x0[1], -32768, 32767); + + return std::array{ + static_cast(x0[0] + fraction * delta0 / scale_factor), + static_cast(x0[1] + fraction * delta1 / scale_factor), + }; + }); } } // namespace AudioInterp diff --git a/src/audio_core/interpolate.h b/src/audio_core/interpolate.h index 19a7b66cb..59f59bc14 100644 --- a/src/audio_core/interpolate.h +++ b/src/audio_core/interpolate.h @@ -6,6 +6,7 @@ #include #include +#include "audio_core/hle/common.h" #include "common/common_types.h" namespace AudioInterp { @@ -14,31 +15,35 @@ namespace AudioInterp { using StereoBuffer16 = std::vector>; struct State { - // Two historical samples. + /// Two historical samples. std::array xn1 = {}; ///< x[n-1] std::array xn2 = {}; ///< x[n-2] + /// Current fractional position. + u64 fposition = 0; }; /** * No interpolation. This is equivalent to a zero-order hold. There is a two-sample predelay. * @param state Interpolation state. * @param input Input buffer. - * @param rate_multiplier Stretch factor. Must be a positive non-zero value. - * rate_multiplier > 1.0 performs decimation and rate_multipler < 1.0 - * performs upsampling. - * @return The resampled audio buffer. + * @param rate Stretch factor. Must be a positive non-zero value. + * rate > 1.0 performs decimation and rate < 1.0 performs upsampling. + * @param output The resampled audio buffer. + * @param outputi The index of output to start writing to. */ -StereoBuffer16 None(State& state, const StereoBuffer16& input, float rate_multiplier); +void None(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output, + size_t& outputi); /** * Linear interpolation. This is equivalent to a first-order hold. There is a two-sample predelay. * @param state Interpolation state. * @param input Input buffer. - * @param rate_multiplier Stretch factor. Must be a positive non-zero value. - * rate_multiplier > 1.0 performs decimation and rate_multipler < 1.0 - * performs upsampling. - * @return The resampled audio buffer. + * @param rate Stretch factor. Must be a positive non-zero value. + * rate > 1.0 performs decimation and rate < 1.0 performs upsampling. + * @param output The resampled audio buffer. + * @param outputi The index of output to start writing to. */ -StereoBuffer16 Linear(State& state, const StereoBuffer16& input, float rate_multiplier); +void Linear(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output, + size_t& outputi); } // namespace AudioInterp -- cgit v1.2.3 From d7459354f58d1b71fc0c5ec48de9242e6a2fd00c Mon Sep 17 00:00:00 2001 From: Subv Date: Mon, 25 Sep 2017 13:06:42 -0500 Subject: Audio: Use std::deque instead of std::vector for the audio buffer type (StereoBuffer16). The current code inserts and deletes elements from the beginning of the audio buffer, which is very inefficient in an std::vector. Profiling was done using VisualStudio2017's Performance Analyzer in Super Mario 3D Land. Before this change: AudioInterp::Linear had 14.14% of the runtime (inclusive) and most of that time was spent in std::vector's insert implementation. After this change: AudioInterp::Linear has 0.36% of the runtime (inclusive) --- src/audio_core/codec.cpp | 4 +++- src/audio_core/codec.h | 4 ++-- src/audio_core/hle/source.h | 2 +- src/audio_core/interpolate.cpp | 2 +- src/audio_core/interpolate.h | 4 ++-- 5 files changed, 9 insertions(+), 7 deletions(-) (limited to 'src/audio_core') diff --git a/src/audio_core/codec.cpp b/src/audio_core/codec.cpp index 7a3bd7eb3..6fba9fdae 100644 --- a/src/audio_core/codec.cpp +++ b/src/audio_core/codec.cpp @@ -117,7 +117,9 @@ StereoBuffer16 DecodePCM16(const unsigned num_channels, const u8* const data, ret[i].fill(sample); } } else { - std::memcpy(ret.data(), data, sample_count * 2 * sizeof(u16)); + for (size_t i = 0; i < sample_count; ++i) { + std::memcpy(&ret[i], data + i * sizeof(s16) * 2, 2 * sizeof(s16)); + } } return ret; diff --git a/src/audio_core/codec.h b/src/audio_core/codec.h index 2b0c395e6..877b2202d 100644 --- a/src/audio_core/codec.h +++ b/src/audio_core/codec.h @@ -5,13 +5,13 @@ #pragma once #include -#include +#include #include "common/common_types.h" namespace Codec { /// A variable length buffer of signed PCM16 stereo samples. -using StereoBuffer16 = std::vector>; +using StereoBuffer16 = std::deque>; /// See: Codec::DecodeADPCM struct ADPCMState { diff --git a/src/audio_core/hle/source.h b/src/audio_core/hle/source.h index ccb7f064f..c4d2debc2 100644 --- a/src/audio_core/hle/source.h +++ b/src/audio_core/hle/source.h @@ -108,7 +108,7 @@ private: u32 current_sample_number = 0; u32 next_sample_number = 0; - std::vector> current_buffer; + AudioInterp::StereoBuffer16 current_buffer; // buffer_id state diff --git a/src/audio_core/interpolate.cpp b/src/audio_core/interpolate.cpp index 16e68bc5c..83573d772 100644 --- a/src/audio_core/interpolate.cpp +++ b/src/audio_core/interpolate.cpp @@ -47,7 +47,7 @@ static void StepOverSamples(State& state, StereoBuffer16& input, float rate, state.xn1 = input[inputi + 1]; state.fposition = fposition - inputi * scale_factor; - input.erase(input.begin(), input.begin() + inputi + 2); + input.erase(input.begin(), std::next(input.begin(), inputi + 2)); } void None(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output, diff --git a/src/audio_core/interpolate.h b/src/audio_core/interpolate.h index 59f59bc14..8dff6111a 100644 --- a/src/audio_core/interpolate.h +++ b/src/audio_core/interpolate.h @@ -5,14 +5,14 @@ #pragma once #include -#include +#include #include "audio_core/hle/common.h" #include "common/common_types.h" namespace AudioInterp { /// A variable length buffer of signed PCM16 stereo samples. -using StereoBuffer16 = std::vector>; +using StereoBuffer16 = std::deque>; struct State { /// Two historical samples. -- cgit v1.2.3 From a13ab958cbba75bc9abd1ca50f3030a10a75784e Mon Sep 17 00:00:00 2001 From: Huw Pascoe Date: Wed, 27 Sep 2017 00:26:09 +0100 Subject: Fixed type conversion ambiguity --- src/audio_core/hle/source.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/audio_core') diff --git a/src/audio_core/hle/source.cpp b/src/audio_core/hle/source.cpp index de4e88cae..c12287700 100644 --- a/src/audio_core/hle/source.cpp +++ b/src/audio_core/hle/source.cpp @@ -264,7 +264,7 @@ void Source::GenerateFrame() { break; } } - state.next_sample_number += frame_position; + state.next_sample_number += static_cast(frame_position); state.filters.ProcessFrame(current_frame); } -- cgit v1.2.3